Similarity-Driven Transfer Learning for Short-Term Building Energy Consumption Forecasting

When considering the use of renewable energy in buildings, accurate short-term energy consumption prediction is critical for achieving a balance between building energy supply and demand. While multi-source transfer learning has improved prediction accuracy, existing studies predominantly focus on total consumption, overlooking the significance of sub-metered energy characteristics. Furthermore, the effectiveness of similarity metrics in such predictions remains underexplored. Therefore, this study introduced a novel multi-source transfer learning model that integrates similarity measurement and sub-metering. This research offers practical insights for optimizing energy consumption predictions in commercial buildings, supports refined energy management strategies and contributes to the development of sustainable, low-carbon buildings.